Perimeter environment laser radar bird's-eye view perspective grid map construction method

Abstract

The application provides a kind of perimeter environment laser radar bird's-eye view perspective grid map construction method, comprising: according to the predefined grid unit size, the laser point cloud frame is discretized, the point cloud set falling into each grid unit is obtained and the point cloud quantity is counted;The point cloud set in the grid unit is vertically layered and noise filtered using DBSCAN algorithm;For the grid unit with effective point cloud quantity greater than 0, the cumulative horizontal centroid coordinates of the grid unit, the cumulative maximum height and the cumulative minimum height of each layer in the vertical direction are calculated and updated;The average point cloud quantity of each grid unit and the interframe mean square deviation are counted and recorded in the grid map together with the cumulative horizontal centroid coordinates, the number of vertical direction layers, the cumulative maximum height and the cumulative minimum height of each layer;The application can eliminate the composition error caused by the ranging noise of the laser radar itself and eliminate the intrusion target missing report problem.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of intelligent security and protection, and in particular to a method for constructing a bird's-eye view grid map of a perimeter environment by using a laser radar. BACKGROUND

[0002] Perimeter intrusion target detection is an important field of intelligent security and protection industry, which can quickly detect and effectively warn the malicious intrusion or climbing of ground targets such as personnel and vehicles in areas with high security requirements such as airports, borders, prisons, etc. Early perimeter intrusion target detection mostly uses electromagnetic microwave shielding, laser shielding, electromagnetic field change, vibration or tension change, etc. However, all of them face the challenge of high false alarm rate under extreme weather conditions such as rain, snow and strong wind.

[0003] As a new type of sensor introduced in the field of autonomous driving, laser radar has three-dimensional ranging and point cloud scanning capabilities, which can obtain the distance, orientation and height of the target relative to the laser radar, and is commonly used for the construction of high-precision maps and the three-dimensional detection of various targets on the road. Through multi-modal fusion with multiple cameras, it can achieve accurate positioning, recognition and trajectory tracking of road targets.

[0004] However, when used for perimeter intrusion detection, laser radar is usually fixedly deployed in a location, facing many challenges such as fluctuation of perimeter terrain and variability of vegetation features in the wild environment. The laser point cloud mapping algorithm in the field of autonomous driving or robotics for structured road environment is mainly aimed at scenarios where the laser radar moves, with high complexity and the need for post-editing of map data. On the other hand, the static environment on the road is relatively simple and cannot be directly applied to complex perimeter environments.

[0005] To solve the above problems, previous existing solutions have proposed a perimeter environment pre-mapping method based on an elevation grid map EGM. However, when the bird's-eye view grid map is discretized and down-sampled, the ranging noise of the laser radar itself causes part of the point cloud returned by large background objects to randomly fall into adjacent blank grid cells, causing errors in mapping and false positives in target detection.

[0006] In addition, the elevation grid map EGM only maintains the elevation information of each grid cell. For background objects such as large trees that stand above the undulating ground surface, even if an intrusion target passes through the grid cell, the elevation value of the grid cell will not change because it is still the height of the background object, resulting in the inability to detect the intrusion target. Although the number of point clouds falling into the grid cell can be counted to some extent to reduce the false alarm rate of the intrusion target, it cannot completely eliminate the problem of missing the intrusion target.

[0007] The bird's-eye view grid map is also known as the BGM (Bird's eye view Grid Map). Summary of the Invention

[0008] To address at least one technical problem in the prior art, this invention provides a method for constructing a perimeter environment lidar bird's-eye view grid map. By calculating the cumulative horizontal centroid coordinates of the grid cells as mapping data for the perimeter environment, it eliminates mapping errors caused by lidar ranging noise and false alarms during target detection. Furthermore, by applying the DBSCAN algorithm to the point cloud set within the grid cells for vertical layering and noise filtering, and recording the cumulative maximum and minimum heights of each layer, it achieves accurate detection of intruding targets located under large trees in densely vegetated environments. To achieve the above technical objectives, the technical solution adopted in this invention is as follows:

[0009] This invention provides a method for constructing a perimeter environment lidar bird's-eye view grid map, comprising the following steps:

[0010] Step S10: Initialize the average number of point clouds and the inter-frame mean square error of the number of point clouds in each grid cell of the bird's-eye view raster map to be constructed, the horizontal centroid coordinates (x, y) of the grid cell, the number of layers in the vertical direction of each grid cell, and the maximum and minimum heights of each layer.

[0011] Step S20: Perform range filtering on each received laser point cloud frame according to the detection area range;

[0012] Step S30: Discretize the laser point cloud frame according to the predefined grid unit size to obtain the point cloud set falling into each grid unit and count the number of point clouds.

[0013] Step S40: Determine whether the number of point clouds in the grid cell is greater than the specified threshold for the number of point clouds. If not, set the number of valid point clouds in the grid cell to zero and proceed to step S60. If yes, proceed to the next step.

[0014] Step S50: The point cloud set within the grid cell is layered vertically and noise filtered along the Z-axis using the DBSCAN algorithm to obtain the number of vertical layers, the maximum and minimum heights of each layer, and the effective point cloud set falling into the grid cell after noise reduction, and the number of effective point clouds is counted.

[0015] Step S60: Add a new historical point cloud count record for the current laser point cloud frame and store the effective point cloud count of all grid cells;

[0016] Step S70: For raster cells with a valid point cloud quantity greater than 0, calculate and update the cumulative horizontal centroid coordinates, the cumulative maximum height and the cumulative minimum height of each layer in the vertical direction of the raster cell.

[0017] Step S80: When the number of received laser point cloud frames exceeds the specified threshold for the number of point cloud frames, based on the historical point cloud count records, the average number of point cloud units and the inter-frame root mean square error are calculated and combined with the cumulative horizontal centroid coordinates and the number of layers in the vertical direction. The cumulative maximum height and cumulative minimum height of each layer are recorded together in the grid diagram; the mapping is complete.

[0018] If the number of laser point cloud frames does not exceed the specified threshold for the number of point cloud frames, return to step S20.

[0019] Furthermore, during initialization, the average number of point clouds per grid cell is set to 0, the inter-frame mean squared error of the number of point clouds is set to 0, the horizontal centroid coordinates (x, y) of the grid cell are set to (0, 0), the number of layers in the vertical direction of each grid cell is set to 0, the maximum height of each layer is set to 0, and the minimum height is set to 1.

[0020] Furthermore, in step S20, the laser point cloud frame is subjected to range filtering based on the pre-set three-dimensional range of the perimeter defense zone, namely its length, width, and height.

[0021] Furthermore, in step S40, the specified threshold for the number of point clouds is configured as 3.

[0022] Furthermore, in step S50, when using the DBSCAN algorithm for vertical layering and noise filtering, the vertical distance threshold of the DBSCAN algorithm is set to 0.6m, and the minimum number of points is set to 3.

[0023] Further, step S70 specifically includes:

[0024] Step S701: Traverse the set of valid point clouds falling into the grid cells and calculate the horizontal centroid coordinates (x, y) of the grid cells, as shown in formula (1):

[0025]

[0026] Where N represents the number of valid point clouds falling into the grid cells in the current laser point cloud frame, and x i ,y i The x-coordinate represents the coordinate of the i-th point in the grid cell. c ,y c Indicates the coordinates of the center point of the grid cell;

[0027] Step S702: Calculate the cumulative horizontal centroid coordinates (x, y) of the grid cells. a ,y a As shown in formula (2):

[0028]

[0029] Where, N a A variable to store the cumulative number of point clouds up to the present;

[0030] Step S703: Based on the maximum height h of each layer of the current laser point cloud frame raster unit obtained in step S50... max and minimum height h min Calculate and update the cumulative maximum height h of each layer. a,max and cumulative minimum height h a,min As shown in formula (3):

[0031]

[0032] Further, step S80 specifically includes:

[0033] Step S801: Based on the historical point cloud quantity records corresponding to each laser point cloud frame, the average point cloud quantity of each grid unit is statistically obtained, the inter-frame mean square error of the point cloud quantity is calculated, and a set of data is recorded as a bird's-eye view grid map.

[0034] Step S802: Record the cumulative horizontal centroid coordinates, the number of vertical layers, and the cumulative maximum and minimum heights of each layer as another set of data for the bird's-eye view raster map.

[0035] Furthermore, in step S80, the threshold for the number of point cloud frames is configured to be 20.

[0036] The beneficial effects of the technical solution provided by the embodiments of the present invention are as follows:

[0037] 1) Regarding the issue that when discretizing and downsampling point cloud frames in a bird's-eye view grid map, the ranging noise of the lidar itself causes some point clouds returned by large background objects to randomly fall into adjacent blank grid cells, since the ranging noise of lidar is generally no greater than 0.1m, by introducing a horizontal centroid in the static map, based on the deviation of the horizontal centroid from the center point of the grid cell (the points generated by ranging noise are mostly located at the edge of the grid cell, and the deviation from the center point of the grid cell is large), it is possible to effectively filter out false alarm events of intrusion targets caused by lidar ranging noise.

[0038] 2) Considering the undulating terrain and dense vegetation of the perimeter environment, the DBSCAN algorithm is used to perform vertical layering and noise filtering on the point cloud set within the grid cell. The cumulative maximum and minimum heights of each layer are recorded. This effectively filters out point cloud noise caused by severe weather conditions such as rain, snow, and fog, reducing the false alarm rate. Furthermore, it effectively maps the perimeter environment of background objects at a certain height above the undulating surface, such as tree canopies, reducing the probability of missed detections of intrusion targets in these scenarios. Since the number of point clouds within a grid cell is much smaller than the number of point clouds in a laser point cloud frame, and DBSCAN clustering is performed only along the vertical direction, the computational complexity of the DBSCAN algorithm in this application is much lower than that of the traditional DBSCAN algorithm applied to the entire laser point cloud frame. Attached Figure Description

[0039] Figure 1 This is a flowchart of the raster map construction method in an embodiment of the present invention. Detailed Implementation

[0040] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0041] In the description of the embodiments of the present invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the present invention. In addition, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0042] In the description of the embodiments of the present invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can also refer to the internal connection of two components; and they can refer to a wireless connection or a wired connection. Those skilled in the art can understand the specific meaning of the above terms in the present invention based on the specific circumstances.

[0043] Furthermore, the technical features involved in the different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0044] like Figure 1 As shown in the figure, this invention proposes a method for constructing a perimeter environment lidar bird's-eye view grid map, including the following steps:

[0045] Step S10: Initialize the average number of point clouds and the inter-frame mean square error of the number of point clouds in each grid cell of the bird's-eye view raster map to be constructed, the horizontal centroid coordinates (x, y) of the grid cell, the number of layers in the vertical direction of each grid cell, and the maximum and minimum heights of each layer.

[0046] In one embodiment, during initialization, the average number of point clouds per grid cell is set to 0, the inter-frame mean squared error of the number of point clouds is set to 0, the horizontal centroid coordinates (x, y) of the grid cell are set to (0, 0), the number of layers in the vertical direction of each grid cell is set to 0, the maximum height of each layer is set to 0, and the minimum height is set to 1.

[0047] In this article, the number of point clouds refers to the number of points in the point cloud; this is hereby clarified.

[0048] Step S20: Perform range filtering on each received laser point cloud frame according to the detection area range;

[0049] In one embodiment, the laser point cloud frame is filtered based on the pre-set three-dimensional range of the perimeter defense zone, namely length, width and height, to filter out point clouds outside the defense zone range and reduce processing complexity.

[0050] Step S30: Discretize the laser point cloud frame according to the predefined grid unit size to obtain the point cloud set falling into each grid unit and count the number of point clouds.

[0051] In one embodiment, the grid cell size is predefined as 0.5m x 0.5m or less than 0.5m x 0.5m;

[0052] Step S40: Determine whether the number of point clouds in the grid cell is greater than the specified threshold for the number of point clouds. If not, set the number of valid point clouds in the grid cell to zero and proceed to step S60. If yes, proceed to the next step.

[0053] In one embodiment, the point cloud quantity threshold is configured as 3;

[0054] Step S50: The point cloud set within the grid cell is layered vertically and noise filtered along the Z-axis using the DBSCAN algorithm to obtain the number of vertical layers, the maximum and minimum heights of each layer, and the effective point cloud set falling into the grid cell after noise reduction, and the number of effective point clouds is counted.

[0055] Specifically, given that the side length of the predefined grid cell in the bird's-eye view grid map is usually no more than 0.5m, DBSCAN clustering is only performed on the point cloud set falling into the grid cell in the vertical direction. That is, the distance metric in the DBSCAN algorithm only considers the z-axis value, rather than the (x,y,z) three-dimensional coordinates, which can complete the layering. At the same time, the complexity of the DBSCAN clustering algorithm is further reduced.

[0056] Specifically, when using the DBSCAN algorithm for vertical layering and noise filtering, the vertical distance threshold of the DBSCAN algorithm is set to 0.6m and the minimum number of points is set to 3 to filter out laser noise generated by severe weather such as rain and snow.

[0057] Specifically, based on the vertical layering results of the DBSCAN algorithm, the maximum and minimum heights of each layer are recorded. Since the DBSCAN algorithm itself has noise filtering characteristics, the abnormal points returned by the LiDAR will not cause abrupt changes in the maximum and minimum height values.

[0058] The number of point clouds remaining after the DBSCAN algorithm filters out outliers is called the effective point cloud number.

[0059] Step S60: Add a new historical point cloud count record for the current laser point cloud frame and store the effective point cloud count of all grid cells;

[0060] Step S70: For raster cells with a valid point cloud count greater than 0, calculate and update the cumulative horizontal centroid coordinates, the cumulative maximum height, and the cumulative minimum height of each layer in the vertical direction for the raster cell; specifically including:

[0061] Step S701: Traverse the set of valid point clouds falling into the grid cells and calculate the horizontal centroid coordinates (x, y) of the grid cells, as shown in formula (1):

[0062]

[0063] Where N represents the number of valid point clouds falling into the grid cells in the current laser point cloud frame, and x i ,y i The x-coordinate represents the coordinate of the i-th point in the grid cell. c ,y c Indicates the coordinates of the center point of the grid cell;

[0064] Step S702: Calculate the cumulative horizontal centroid coordinates (x, y) of the grid cells. a ,y a As shown in formula (2):

[0065]

[0066] Where, Na A variable to store the cumulative number of point clouds up to the present;

[0067] Step S703: Based on the maximum height h of each layer of the current laser point cloud frame raster unit obtained in step S50... max and minimum height h min Calculate and update the cumulative maximum height h of each layer. a,max and cumulative minimum height h a,min As shown in formula (3):

[0068]

[0069] It should be noted that for large background objects such as trees that stand vertically on undulating ground, due to the large number of point clouds, the laser point cloud frame will not sometimes be divided into multiple layers and sometimes only one layer. Therefore, there is no need to deal with the problem of different vertical layers between different laser point cloud frames.

[0070] Step S80: When the number of received laser point cloud frames exceeds the specified threshold for the number of point cloud frames, based on the historical point cloud count records, the average number of point cloud units and the inter-frame root mean square error are calculated and combined with the cumulative horizontal centroid coordinates and the number of layers in the vertical direction. The cumulative maximum height and cumulative minimum height of each layer are recorded together in the grid diagram; the mapping is complete.

[0071] When the number of laser point cloud frames does not exceed the specified threshold for the number of point cloud frames, return to step S20;

[0072] Specifically, it includes:

[0073] Step S801: Based on the historical point cloud quantity records corresponding to each laser point cloud frame, the average point cloud quantity of each grid unit is statistically obtained, the inter-frame mean square error of the point cloud quantity is calculated, and a set of data is recorded as a bird's-eye view grid map.

[0074] Step S802: Record the cumulative horizontal centroid coordinates, the number of vertical layers, and the cumulative maximum and minimum heights of each layer as another set of data for the bird's-eye view raster map;

[0075] In this embodiment, the threshold for the number of point cloud frames is configured as 20.

[0076] Finally, it should be noted that the above specific embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A perimeter environment laser radar bird's eye view grid map construction method, characterized in that, The method comprises the following steps: Step S10, initializing the average point cloud quantity of each grid cell in the bird's eye view grid map to be constructed, the inter-frame mean square deviation of the point cloud quantity, the horizontal centroid coordinates (x, y) of the grid cell, the number of layers in the vertical direction of each grid cell, the maximum height and the minimum height of each layer; Step S20, range filtering is performed on each received laser point cloud frame according to the detection area range; Step S30, discretization is performed on the laser point cloud frame according to the predefined grid cell size, to obtain the point cloud set falling into each grid cell and count the point cloud quantity; Step S40, it is judged whether the point cloud quantity in the grid cell is greater than the specified point cloud quantity threshold, if not, the effective point cloud quantity in the grid cell is directly set to zero, and step S60 is entered, if yes, the next step is performed; Step S50, the point cloud set in the grid cell is subjected to vertical layering and noise filtering along the Z axis by using the DBSCAN algorithm, to obtain the number of layers in the vertical direction, the maximum height and the minimum height of each layer, and obtain the effective point cloud set falling into the grid cell after noise reduction and count the effective point cloud quantity; Step S60, a historical point cloud quantity record is added for the current laser point cloud frame, and the effective point cloud quantities of all grid cells are stored; Step S70, for the grid cell with the effective point cloud quantity greater than 0, the accumulated horizontal centroid coordinates of the grid cell, the accumulated maximum height and the accumulated minimum height of each layer in the vertical direction are calculated and updated; Step S80, when the number of received laser point cloud frames exceeds the specified point cloud frame quantity threshold, the average point cloud quantity and the inter-frame mean square deviation of each grid cell are counted according to the historical point cloud quantity record, and the accumulated horizontal centroid coordinates, the number of layers in the vertical direction, the accumulated maximum height and the accumulated minimum height of each layer are recorded in the grid map together; The mapping is completed; When the number of laser point cloud frames does not exceed the specified point cloud frame quantity threshold, returning to step S20; The step S70 specifically comprises: Step S701, traversing the effective point cloud set falling into the grid cell, calculating the horizontal centroid coordinates (x, y) of the grid cell, as shown in formula (1): wherein N represents the number of valid point clouds in the current laser point cloud frame falling into the grid unit, x i ,y i represents the coordinates of the i-th point in the grid unit, x c ,y c represents the coordinates of the center point of the grid unit; Step S702, calculate the accumulated horizontal centroid coordinates (x a ,y a ) of the grid unit, as shown in formula (2): N a is a variable for storing the cumulative point cloud quantity up to the present; Step S703, according to the maximum height h and the minimum height h of each layer of the current laser point cloud frame grid unit obtained in step S50 max , the accumulated maximum height h and the accumulated minimum height h of each layer are calculated and updated min , as shown in formula (3): a,max a,min ​​ 2. The perimeter environment laser radar bird's eye view grid map construction method according to claim 1, wherein During initialization, the average point cloud quantity of each grid cell is set to 0, the inter-frame mean square deviation of the point cloud quantity is set to 0, the horizontal centroid coordinates (x, y) of the grid cell are set to (0, 0), the number of layers in the vertical direction of each grid cell is set to 0, and the maximum height and the minimum height of each layer are set to 0 and 1 respectively.

3. The perimeter environment laser radar bird's eye view grid map construction method according to claim 1, wherein In step S20, the laser point cloud frame is subjected to range filtering according to the pre-set three-dimensional range of the perimeter defense area, i.e. the length, the width and the height.

4. The perimeter environment laser radar bird's eye view grid map construction method according to claim 1, wherein In step S40, the specified point cloud quantity threshold is configured to be 3.

5. The perimeter environment laser radar bird's eye view grid map construction method according to claim 1, wherein In step S50, when the vertical direction is layered and noise filtering is performed by using the DBSCAN algorithm, the vertical distance threshold of the DBSCAN algorithm is set to 0.6 m, and the minimum point quantity is set to 3. 6.The perimeter environment laser radar bird's eye view grid map construction method of claim 1, characterized in that, The step S80 specifically comprises: Step S801, according to the historical point cloud quantity record corresponding to each laser point cloud frame, the average point cloud quantity of each grid unit is obtained by statistics, the inter-frame mean square deviation of the point cloud quantity is calculated; and a group of data of the bird's eye view grid map is recorded; Step S802, record the cumulative horizontal centroid coordinates, the number of vertical direction layers, the cumulative maximum height and the cumulative minimum height of each layer, as another group of data of the bird's eye view grid map. 7.The perimeter environment laser radar bird's eye view grid map construction method of claim 1, characterized in that, In the step S80, the point cloud frame quantity specified threshold is configured to be 20.

Images